diff --git a/kmath-memory/src/commonMain/kotlin/kscience/kmath/memory/DeferScope.kt b/kmath-memory/src/commonMain/kotlin/kscience/kmath/memory/DeferScope.kt
new file mode 100644
index 000000000..df5040ee0
--- /dev/null
+++ b/kmath-memory/src/commonMain/kotlin/kscience/kmath/memory/DeferScope.kt
@@ -0,0 +1,7 @@
+package kscience.kmath.memory
+
+public expect class DeferScope {
+    public inline fun defer(crossinline block: () -> Unit)
+}
+
+public expect inline fun <R> withDeferScope(block: DeferScope.() -> R): R
diff --git a/kmath-memory/src/jsMain/kotlin/kscience/kmath/memory/DeferScopeJS.kt b/kmath-memory/src/jsMain/kotlin/kscience/kmath/memory/DeferScopeJS.kt
new file mode 100644
index 000000000..aeb699a39
--- /dev/null
+++ b/kmath-memory/src/jsMain/kotlin/kscience/kmath/memory/DeferScopeJS.kt
@@ -0,0 +1,30 @@
+package kscience.kmath.memory
+
+private typealias Deferred = () -> Unit
+
+public actual class DeferScope {
+    @PublishedApi
+    internal val deferred: MutableList<Deferred> = mutableListOf()
+
+    @PublishedApi
+    internal fun executeAllDeferred() {
+        deferred.forEach(Deferred::invoke)
+        deferred.clear()
+    }
+
+    public actual inline fun defer(crossinline block: () -> Unit) {
+        deferred += {
+            try {
+                block()
+            } catch (ignored: Throwable) {
+            }
+        }
+    }
+}
+
+public actual inline fun <R> withDeferScope(block: DeferScope.() -> R): R {
+    val ds = DeferScope()
+    val r = ds.block()
+    ds.executeAllDeferred()
+    return r
+}
diff --git a/kmath-memory/src/jvmMain/kotlin/kscience/kmath/memory/DeferScopeJvm.kt b/kmath-memory/src/jvmMain/kotlin/kscience/kmath/memory/DeferScopeJvm.kt
new file mode 100644
index 000000000..aeb699a39
--- /dev/null
+++ b/kmath-memory/src/jvmMain/kotlin/kscience/kmath/memory/DeferScopeJvm.kt
@@ -0,0 +1,30 @@
+package kscience.kmath.memory
+
+private typealias Deferred = () -> Unit
+
+public actual class DeferScope {
+    @PublishedApi
+    internal val deferred: MutableList<Deferred> = mutableListOf()
+
+    @PublishedApi
+    internal fun executeAllDeferred() {
+        deferred.forEach(Deferred::invoke)
+        deferred.clear()
+    }
+
+    public actual inline fun defer(crossinline block: () -> Unit) {
+        deferred += {
+            try {
+                block()
+            } catch (ignored: Throwable) {
+            }
+        }
+    }
+}
+
+public actual inline fun <R> withDeferScope(block: DeferScope.() -> R): R {
+    val ds = DeferScope()
+    val r = ds.block()
+    ds.executeAllDeferred()
+    return r
+}
diff --git a/kmath-memory/src/nativeMain/kotlin/kscience/kmath/memory/DeferScopeNative.kt b/kmath-memory/src/nativeMain/kotlin/kscience/kmath/memory/DeferScopeNative.kt
new file mode 100644
index 000000000..6809b18b3
--- /dev/null
+++ b/kmath-memory/src/nativeMain/kotlin/kscience/kmath/memory/DeferScopeNative.kt
@@ -0,0 +1,7 @@
+package kscience.kmath.memory
+
+import kotlinx.cinterop.memScoped
+
+public actual typealias DeferScope = kotlinx.cinterop.DeferScope
+
+public actual inline fun <R> withDeferScope(block: DeferScope.() -> R): R = memScoped(block)
diff --git a/kmath-torch/src/nativeMain/kotlin/kscience.kmath.torch/TorchTensor.kt b/kmath-torch/src/nativeMain/kotlin/kscience.kmath.torch/TorchTensor.kt
index cc41359a2..3e2ebd562 100644
--- a/kmath-torch/src/nativeMain/kotlin/kscience.kmath.torch/TorchTensor.kt
+++ b/kmath-torch/src/nativeMain/kotlin/kscience.kmath.torch/TorchTensor.kt
@@ -2,13 +2,14 @@ package kscience.kmath.torch
 
 
 import kscience.kmath.structures.TensorStructure
+import kscience.kmath.memory.DeferScope
 
 import kotlinx.cinterop.*
 import kscience.kmath.ctorch.*
 
 
 public sealed class TorchTensor<T> constructor(
-    internal val scope: DeferScope,
+    public val scope: DeferScope,
     internal val tensorHandle: COpaquePointer
 ) : TensorStructure<T>() {
     init {
@@ -79,7 +80,7 @@ public sealed class TorchTensorOverField<T> constructor(
     scope: DeferScope,
     tensorHandle: COpaquePointer
 ) : TorchTensor<T>(scope, tensorHandle) {
-    internal var requiresGrad: Boolean
+    public var requiresGrad: Boolean
         get() = requires_grad(tensorHandle)
         set(value) = requires_grad_(tensorHandle, value)
 }
diff --git a/kmath-torch/src/nativeMain/kotlin/kscience.kmath.torch/TorchTensorAlgebra.kt b/kmath-torch/src/nativeMain/kotlin/kscience.kmath.torch/TorchTensorAlgebra.kt
index dc66813de..00d123c4a 100644
--- a/kmath-torch/src/nativeMain/kotlin/kscience.kmath.torch/TorchTensorAlgebra.kt
+++ b/kmath-torch/src/nativeMain/kotlin/kscience.kmath.torch/TorchTensorAlgebra.kt
@@ -2,6 +2,8 @@ package kscience.kmath.torch
 
 
 import kscience.kmath.structures.*
+import kscience.kmath.memory.DeferScope
+import kscience.kmath.memory.withDeferScope
 
 import kotlinx.cinterop.*
 import kscience.kmath.ctorch.*
@@ -591,23 +593,23 @@ public class TorchTensorIntAlgebra(scope: DeferScope) :
 }
 
 public inline fun <R> TorchTensorRealAlgebra(block: TorchTensorRealAlgebra.() -> R): R =
-    memScoped { TorchTensorRealAlgebra(this).block() }
+    withDeferScope { TorchTensorRealAlgebra(this).block() }
 
 public inline fun <R> TorchTensorFloatAlgebra(block: TorchTensorFloatAlgebra.() -> R): R =
-    memScoped { TorchTensorFloatAlgebra(this).block() }
+    withDeferScope { TorchTensorFloatAlgebra(this).block() }
 
 public inline fun <R> TorchTensorLongAlgebra(block: TorchTensorLongAlgebra.() -> R): R =
-    memScoped { TorchTensorLongAlgebra(this).block() }
+    withDeferScope { TorchTensorLongAlgebra(this).block() }
 
 public inline fun <R> TorchTensorIntAlgebra(block: TorchTensorIntAlgebra.() -> R): R =
-    memScoped { TorchTensorIntAlgebra(this).block() }
+    withDeferScope { TorchTensorIntAlgebra(this).block() }
 
-public fun TorchTensorReal.withGrad(block: TorchTensorRealAlgebra.() -> TorchTensorReal): TorchTensorReal {
+public inline fun TorchTensorReal.withGrad(block: TorchTensorRealAlgebra.() -> TorchTensorReal): TorchTensorReal {
     this.requiresGrad = true
     return TorchTensorRealAlgebra(this.scope).block()
 }
 
-public fun TorchTensorFloat.withGrad(block: TorchTensorFloatAlgebra.() -> TorchTensorFloat): TorchTensorFloat {
+public inline fun TorchTensorFloat.withGrad(block: TorchTensorFloatAlgebra.() -> TorchTensorFloat): TorchTensorFloat {
     this.requiresGrad = true
     return TorchTensorFloatAlgebra(this.scope).block()
 }
\ No newline at end of file